Hey Trump,
I'm admin'ing a 42 processor Linux cluster at work that is based on Tyan's dual AMD motherboards (a mix of MP and MPX chipsets). I don't have our benchmarks in front of me at the moment, but I can say that it's super fast. We compiled a custom kernel (2.4.17) with the athlon arcitecture flag thrown. Our applications are also compiled with the athlon arcitecture flag. Your choice of compiler will have an impact on application performance. Our kernel is compiled with GNU, but our apps are compiled with a mix of Intel and Portland Group C compilers. The latest beta Intel compiler performs exceptionally well on Athlon MPs.
Regarding SCSI... only go down that road if you need:
(a) more than 2 drivers per controller channel
(b) super fast access times
The sustained read and write speeds of SCSI aren't stellar enough when compared with ATA100/133 to justify their enormous price tag.
I'd be pretty surprised if the performance of GF2, GF3, and GF4 weren't quite different for OpenGL apps. The lastest nvidia drivers for Linux enhance OpenGL performance and add in some tweaks for GF4. The question is... how many OpenGL apps will you be running on your server? If the answer is 'not much', then save some cash and get a GF2.
In summary, I'd go with dual AMD with IDE because you get great performance at a great price. I'd go with an MPX chipset or one of the Tyan Thunder MP boards without onboard LAN and SCSI -- again to save $$$. Go with GF2 unless you plan on doing a lot of OpenGL gaming. If you wanna do OpenGL or run Windows DirectX games via Code Weavers WINE (*gasp*), then get something juicier (hmm, did i spell that right?).
==Flipper==
|